The OBS instruments used in this study were short-period seismometers (MicrOBS) supplied by the L’Institut Français de Recherche pour l’Exploitation de la Mer. The MicrOBS data logger is stable and efficiently recorded ground motion in the frequency band of 1–100 Hz with a peak response at 4.5 Hz (Auffret et al. 2004). Seismic airgun shots proceeded OBS deployments to relocate the seafloor OBS stations and to retrieve the recording directions of the two horizontal components (Chang et al. 2008).
Prior to the outer trench slope sequence that occurred in 2006, we had already deployed Network 2005 at the lower slope of the frontal wedge in 2005 to detect the seismicity in the trench area. We used a standard seismic program, HYPOINVERSE2000 (Klein 2002), to determine the earthquake hypocenters detected by using the OBS networks. The initial velocity model (Figure 4) is based on Eakin et al. (2014: Profile T2, Figure 11). With hypocentral determination with the initial velocity model, we refined the seismic velocity layer by layer from top to bottom by minimizing the travel-time residuals and iteratively relocating the earthquakes. The program VELEST (Kissling et al. 1994) was employed to retrieve the optimized velocity model with each OBS experiment. Events with a travel-time residual larger than 0.3 s, approximately 5% of the detections, were removed. Ultimately, approximately 400 and 2400 local earthquakes were retrieved from the campaign records of Networks 2005 and 2006, respectively. We hereafter refer to these two earthquake groups as C2005 and C2006, respectively. The hypocentral parameters of C2005 and C2006 are listed in Table S2. The temporal evolution of these groups is presented in Figure 2. Figure 3 presents a map view of the seismicity density (per km2) at various focal depth levels. In this study, zero focal depth indicates sea level.
In accordance with the spatial range of focal depths, the retrieved seismic models defined the velocity structures approximately 25 and 35 km below the seafloor for Networks 2005 and 2006, respectively. Because unconsolidated sediments are distributed across the seafloor, we allowed for, but did not preset, a low-velocity layer in the topmost kilometers when developing the velocity models. Figure 4 presents the layered seismic models generated from C2005 and C2006; Vp/Vs was evaluated in terms of depth. These two seismic velocity models appear to be consistent with oceanic crust rather than continental crust. However, the velocity model determined using C2005 indicates a structure of dipping downgoing oceanic crust with low-velocity sedimentary rocks in the top layers. The seismic parameters of these two groups of earthquakes are presented in detail as follows.
C2005: Earthquakes in the frontal wedge of the overriding plate
C2005 comprised earthquakes occurring over a period of 5 days (October 7–11). A mainshock of magnitude 4.5 occurred on October 9 (Event 67 in Table S2) and was recorded by Network 2005 at a depth of approximately 9 km. Before this event, microseisms were scattered within the shallow layers at approximately 5 km (Figure 2). The seismicity increased simultaneously with the emergence of the mainshock, and the focal depth distribution of C2005 was dramatically extended to approximately 25 km from sea level (Figure 2). The accurate hypocentral determination reveals that this sequence was distributed in a spatially vertical zone over a depth range of 20 km. This type of spatial pattern is not commonly detected in the subduction zone. In the frontal wedge area, seismicity usually indicates a regime of hydrologic and tectonic processes with active faults. However, the seismogenic zone was recognized by a décollement dividing an upper brittle-fracture-dominated domain overlying a lower, ductile domain. Pore-water geochemical evidence reveals that along-fault flow occurs specifically in the upper brittle domain but is hydrologically isolated from fluids in the underlying footwall sediments (e.g., Cello and Nurr 1988; Tobin et al. 2001; Lin Andrew et al. 2009). The C2005 data indicate vital seismic activity in which the up-thrust faulting emerged at the top crust of the frontal wedge area.
The seismic velocity model constructed with C2005 indicated low velocity for the top layers within 4 km from the seafloor. The P- and S-wave velocities are approximately <3.0 and <1.8 km/s, respectively, and Vp/Vs > 1.80. In particular, in the topmost layer (~1 km) the P- and S-wave velocities are as low as 1.9 and 0.6 km/s, respectively. Vp/Vs for this layer is 1.83. This low-velocity impedance is common in unconsolidated sediments at shallow accretionary prisms. Network 2005 was located in a field of widely distributed gas hydrate in southwest offshore Taiwan (e.g., Liu et al. 2006; Berndt et al. 2019). Thus, a high Vp/Vs at the topmost layer was expected. At the depth of 4 km beneath than the seafloor, the P- and S-velocities are much higher at approximately 5.8 and 3 km/s, respectively. This sharp change in seismic velocity corresponds to the interface between accretionary sediments and hard basement. However, the Vp/Vs ratio remains as large as 1.80 at the upper and lower layers of the sediment–hard rock interface (Figure 4). We considered high fluid content at the interface as well as in the accretionary sediments in the frontal area. Notably, this is the depth level at which where we observed microseisms before the mainshock of Event 67; it is also the focal depth of the larger earthquakes (including the mainshock) of this sequence. We inferred, therefore, that the upper layers, comprising the sedimentary prism and the top crust, have high fluid content, and overpressure of this fluid is responsible for the earthquake sequence beginning on October 9.
Below 9 km, the seismic velocity increases and Vp/Vs decreases until approximately 16 km. At depths below 16 km, the seismic velocity is almost invariant; the P-wave velocity is approximately 7.8–8.0 km/s. However, Vp/Vs increases to 1.75 from 16 to 20 km. This may indicate a change in lithology from oceanic crust to lithospheric mantle. The Moho depth at the frontal wedge site should be approximately 16 km below the sea level, as revealed by the C2005 data.
C2006: Earthquakes in the outer trench slope of the coming plate
Network 2006 was located on the outer trench slope west of the Manila Trench, and earthquake magnitudes in C2006 ranged from 1 to 3.5 without temporal or spatial variation during the experimental period (Figure 2). However, the seismicity rate in C2006 was as high as 500 events per day. Combining our result with the global seismic observations reveals that this bending–extensional sequence lasted for months—at least from the middle to the end of 2006.
The hypocenters in C2006 reveal two fault planes at different depth ranges: a high-angle north–south striking fault plane dipping to the northeast and located from the seafloor to approximately 25 km and a low-angle fault plane lying at approximately 20–35 km (Figure 3). C2006 were the aftershock events of the extensional sequence initialized in the middle of 2006. Both GCMT and AutoBATS indicate a north–south striking normal fault for the large events of this sequence (Figures 1 and S1). We recognize that the high-angle northeastward-dipping fault plane revealed by C2006 is the actual fault plane of this normal fault sequence. However, few clues regarding the formation mechanism for this low-angle fault with extensional deformation of the plate have been obtained. Based on the seismic velocities obtained in C2006, the Moho depth should be approximately 8 km below the seafloor (~12 km below sea level); at and below this depth, the P-wave velocity was as much as 7.6 km/s and invariant. This Moho depth reading is consistent with the previous active seismic surveys in this area (e.g., Wang et al. 2006; Eakin et al. 2014; Lester et al. 2014) In the other words, the low-angle fault plane is deformation in the upper mantle.
Because the northeast SCS contains relict structures from primitive basin spreading, we infer that the low-angle fault in C2006 is the deep detachment fault, which is the lowest bound of the listric normal faults that form in the hyperextension (Reston et al. 1996).
In addition to information about the Moho depth, the seismic velocity model from C2006 also indicates a relatively low lithological strength for the oceanic plate (Figure 4). The P-velocity at the topmost layer is approximately 3.8 km/s, indicating sedimentary rock or hard rock. However, Vp/Vs at these layers is greater than 1.80. This may indicate mechanical weakening or hydro fractures in the outer trench slope area caused by extensional fractures or indicate the filling of the fractures by sediments with high fluid content, or both (e.g., Tobin et al. 2014).
Furthermore, the Vp/Vs ratios from C2006 are lower than those from C2005 at the layers above 20 km, indicating that the dehydration is dramatic as the plate subducts into the mantle.